datacenter
Thuê Nvidia B200.
Blackwell
192GB VRAM
1000W
From $2.79/hr
Per hour
$2.79
Per day
$66.96
Per week
$468.72
Per month
$2009
Provider spread
6 providers ·
up to 58% cheaper at the low end
Cheapest · $2.79/hr on DeepInfra
Median $4.49/hr
Most expensive · $6.69/hr on Lambda Labs
Price history
Daily median across providers.
Loading...
Where to rent it
All providers carrying this GPU.
Buy vs rent
Should you rent or own?
Your usage
Rent on cloud
Per day
—
Per month
—
at $2.79/hr cheapest provider rate
Own on-prem
Electricity per day
—
1000W TDP
Rent breakeven on $40,000 MSRP
— days
· — months
Workloads
Suitable workloads.
AI models that fit
See all 2 →
Run these on this GPU.
- Llama 3.1 405B 2× · fp8
- Kimi K2 4× · fp8
Cloud instances
See all 3 →
Hyperscaler bundles.
Pre-configured on 3 clouds — from $4.99/hr total
($4.99/hr per GPU).
FAQ
Frequently asked.
How is the Nvidia B200 price calculated?
We pull live listings from each provider's public API, take the median hourly rate across active offers, and refresh every hour. The rate shown is the median, so a single low-ball spot offer can't distort the headline.
Why does the Nvidia B200 cost different amounts on different providers?
P2P marketplaces like Vast.ai aggregate offers from individual hosts who set their own rates — bidding pushes prices down. First-party clouds (Lambda, hyperscalers) charge a managed-service premium for support, SLAs, and integrated networking. Decentralized networks (io.net, Akash) settle in tokens, which adds volatility but often the lowest base rate.
Can I really train an LLM on a single Nvidia B200?
Depends on the model size. With 192GB of VRAM you can fine-tune 70B models with QLoRA, or fully fine-tune 13B-class models.
Spot vs on-demand on the Nvidia B200 — which should I rent?
On-demand keeps the same instance until you stop it; spot (or interruptible) is cheaper but the host can reclaim it when a higher-paying job lands. Use on-demand for training runs and anything stateful. Use spot for stateless inference, batch jobs, and experiments where a checkpoint every few minutes is enough to recover.
How is hourly billing measured for the Nvidia B200?
Most providers bill per-second once the instance is running, with a small minimum (often 60 seconds). A handful of first-party clouds round up to the minute. Either way, headline $/hr is the right comparison unit.
Does the region of the host affect the Nvidia B200 price?
Yes — US and EU regions usually carry a premium over LATAM, India, and parts of APAC, especially on first-party clouds. P2P marketplaces hide this behind one global price because supply moves wherever bids exist.
Cloud instance options
Pre-configured instances on hyperscalers.
Whole-instance bundles (GPU + vCPU + RAM + disk) on the major clouds. Per-GPU rate often drops as the count rises. View = spec page · Launch = sign up (affiliate).
Cheapest bundle
$4.99/hr
Lowest $/hr per GPU
$4.99/hr
Providers
3
Instance shapes
3
| Provider | Instance | GPUs | vCPU | RAM | Disk | $/hr | $/hr per GPU | |
|---|---|---|---|---|---|---|---|---|
| gpu_1x_b200 | 1× | — | — | — | $4.99/hr | $4.99/hr | ||
| BM.GPU.B200.8 | 8× | — | — | — | $144.00/hr | $18.00/hr | ||
| p6-b200.48xlarge | 8× | — | — | — | $160.00/hr | $20.00/hr |
Looking for the cheapest rate?
Hyperscaler bundles include managed networking + SLAs. Raw per-GPU rental on P2P marketplaces is typically 3–10× cheaper.
See raw rental rates on the Overview tab →
FAQ
Cloud instances — common questions.
How is the Nvidia B200 price calculated?
We pull live listings from each provider's public API, take the median hourly rate across active offers, and refresh every hour. The rate shown is the median, so a single low-ball spot offer can't distort the headline.
Why does the Nvidia B200 cost different amounts on different providers?
P2P marketplaces like Vast.ai aggregate offers from individual hosts who set their own rates — bidding pushes prices down. First-party clouds (Lambda, hyperscalers) charge a managed-service premium for support, SLAs, and integrated networking. Decentralized networks (io.net, Akash) settle in tokens, which adds volatility but often the lowest base rate.
Can I really train an LLM on a single Nvidia B200?
Depends on the model size. With 192GB of VRAM you can fine-tune 70B models with QLoRA, or fully fine-tune 13B-class models.
Spot vs on-demand on the Nvidia B200 — which should I rent?
On-demand keeps the same instance until you stop it; spot (or interruptible) is cheaper but the host can reclaim it when a higher-paying job lands. Use on-demand for training runs and anything stateful. Use spot for stateless inference, batch jobs, and experiments where a checkpoint every few minutes is enough to recover.
How is hourly billing measured for the Nvidia B200?
Most providers bill per-second once the instance is running, with a small minimum (often 60 seconds). A handful of first-party clouds round up to the minute. Either way, headline $/hr is the right comparison unit.
Does the region of the host affect the Nvidia B200 price?
Yes — US and EU regions usually carry a premium over LATAM, India, and parts of APAC, especially on first-party clouds. P2P marketplaces hide this behind one global price because supply moves wherever bids exist.
AI models
FP8
FP8
Models that run on this GPU.
GPU-count + quantization recommendations covering fine-tuning, inference, and run-it-yourself scenarios on the Nvidia B200.
Llama 3.1 405B
by Meta AI
Required GPUs
2× Nvidia B200
Total VRAM
384 GB
Parameters
405B
Context
128K tokens
Kimi K2
by Moonshot AI
Required GPUs
4× Nvidia B200
Total VRAM
768 GB
Parameters
1000B (32B active)
Context
256K tokens
Renting for inference?
Pair these models with the cheapest provider on the rental table.
See rental rates on the Overview tab →
FAQ
AI models on this GPU.
How is the Nvidia B200 price calculated?
We pull live listings from each provider's public API, take the median hourly rate across active offers, and refresh every hour. The rate shown is the median, so a single low-ball spot offer can't distort the headline.
Why does the Nvidia B200 cost different amounts on different providers?
P2P marketplaces like Vast.ai aggregate offers from individual hosts who set their own rates — bidding pushes prices down. First-party clouds (Lambda, hyperscalers) charge a managed-service premium for support, SLAs, and integrated networking. Decentralized networks (io.net, Akash) settle in tokens, which adds volatility but often the lowest base rate.
Can I really train an LLM on a single Nvidia B200?
Depends on the model size. With 192GB of VRAM you can fine-tune 70B models with QLoRA, or fully fine-tune 13B-class models.
Spot vs on-demand on the Nvidia B200 — which should I rent?
On-demand keeps the same instance until you stop it; spot (or interruptible) is cheaper but the host can reclaim it when a higher-paying job lands. Use on-demand for training runs and anything stateful. Use spot for stateless inference, batch jobs, and experiments where a checkpoint every few minutes is enough to recover.
How is hourly billing measured for the Nvidia B200?
Most providers bill per-second once the instance is running, with a small minimum (often 60 seconds). A handful of first-party clouds round up to the minute. Either way, headline $/hr is the right comparison unit.
Does the region of the host affect the Nvidia B200 price?
Yes — US and EU regions usually carry a premium over LATAM, India, and parts of APAC, especially on first-party clouds. P2P marketplaces hide this behind one global price because supply moves wherever bids exist.